Defect detection - Halcon surface scratch detection

One . Project scenario ：

Extract surface scratch defects

---

Two . Solution ：

1. First, do median filtering on the original image ：

2. Local threshold processing

3.blob Analyze and extract features

3、 ... and .halcon Code

read_image (Image, 'scratchSrc')
mean_image (Image, ImageMean, 7, 7)
dyn_threshold (Image, ImageMean, DarkPixels, 5, 'dark')
connection (DarkPixels, ConnectedRegions)
select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 10, 1000)
union1 (SelectedRegions, RegionUnion)
dilation_circle (RegionUnion, RegionDilation, 3.5)
skeleton (RegionDilation, Skeleton)
connection (Skeleton, Errors)
select_shape (Errors, Scratches, 'area', 'and', 50, 10000)
select_shape (Errors, Dots, 'area', 'and', 1, 50)

Four . Core algorithm analysis

dyn_threshold (OrigImage, ThresholdImage, RegionDynThresh, Offset, LightDark）

Algorithm function ： Use local threshold to segment the image .

OrigImage： Original input image
ThresholdImage： Input image after smoothing
RegionDynThresh： Result output image
Offset: be relative to ThresholdImage The offset of each pixel value of the image
LightDark： Options , Extract light 、 Dark or similar areas ?
The default value is :“light” Options provided : ‘dark’, ‘equal’, ‘light’, ‘not_equal’

Analysis of the original figure shows , The scratch and background in the figure are very small , It is completely impossible to find a suitable global threshold to segment the scratch from the background . Therefore, the local threshold segmentation algorithm is the solution The key to .
The principle is as follows ：
Select the area where the pixel meets the threshold condition from the input image , Make g{o} = g{OrigImage}, g{t} = g{ThresholdImage}. that LightDark= 'dark’ Is the condition of :

g{o} <= g{t} - Offset

OrigImag and ThresholdImage Each pixel is compared one by one , When ThresholdImage Pixel value of minus Offset The offset is greater than or equal to OrigImag Pixel value of , said OrigImag The pixel value of meets the conditional threshold , In pixel values 255 Save to RegionDynThresh in , If the rest are not satisfied, the pixel value 0 Save in RegionDynThresh in .
If LightDark= 'light’ Then the condition is :

g{o} >= g{t} - Offset

c++ Source code implementation effect